Role of Agents in Distributed Network Management: A Review

Transcription

1 Abstract Role of Agents in Distributed Networ Management: A Review Atul Mishra, A.K.Sharma Department of Comp. Engg. YMCA University of Science and Technology, Faridabad, India [email protected], [email protected] Today s networ management systems, mostly based on SNMP or CMIP management protocols, suffer from scalability and flexibility problems as it involve the transmission of large amount of management data towards the centralized management station for processing. In the event of networ stress (e.g. generation of lots of fault data or performance monitoring data) it overloads the management station. A distributed paradigm is a viable solution to perform management functions when networ start growing significantly. In this sense, use of mobile agents offer many possibilities for designing the next generation of distributed networ management systems. They decentralize processing and control thereby reducing the management traffic around the management station and distribute processing load. This paper discusses and analyzes the ey advantages and application of mobile agents in the distributed networ management systems. Keywords: Mobile agents, Networ Management, Distributed, SNMP, Scalability 1. Introduction Modern day telecommunication industry is facing ever increasing demand for more sophisticated services, higher quality and shorter times to maret. Networ operators and service providers in the telecommunications industry must meet these demands at a commercially viable cost and should have plans in place for any unforeseen change. This flexibility must be achieved in an environment that is complicated by distributed, often mobile, data, resources, service access and control [1], especially when these networs are growing in size and complexity. Moreover, varied technologies, such as SONET, ATM, Ethernet, DWDM etc., present at different layers of the Access, Metro and Core (long haul) sections of the networ, have contributed to the complexity in terms of their own framing and protocol structures. Thus, controlling and managing the traffic in these networs is a challenging tas. Networ management system is a fundamental facilitator that typically gathers and analyzes huge amount of data from the networ and maes decisions thereof for various functional areas of a management system. The real-time requirements for various types of traffic Vol 1, No 2 (October 2010) IJoAT 284

2 are also set instantly. Existing management models traditionally adopt a centralized, client/server (CS) approach wherein the management application, with the help of a manager, acting as clients, periodically accesses the data collected by a set of software modules, agents, placed on networ devices by using an appropriate protocol as shown in figure 1. NMS NMS NMS NMS Applications & Management Station MANAGER Client Req/Res Event s AGENT AGENT AGENT Server Managed Nodes MIB MIB MIB Figure 1. Centralized Manager/Agent interaction The functionality of both managers and the agents where managers act as client and the agents as a server is statically defined at design time. Traditionally, the Networ Management (NM) scene has been dominated by the Internet Engineering Tas Force (IETF) simple networ management protocol (SNMP) for data networs [2], and the OSI common management information protocol (CMIP) for telecommunication networ [3], which are typically designed according to a centralized model and hence suffer from lac of distribution, a low degree of flexibility and fault tolerance [4]. They also require networ operators at NMS level to mae real-time decisions and manually find solutions for the series of problems in the networ. These networ management systems deal only with data gathering and reporting methods, which in general involve substantial transmission of management data thereby consuming a lot of bandwidth and computational overhead. Moreover it also causes a considerable strain on the networ and significant traffic jam at the manager host [5]. Besides this centralized management activities are limited in their capability as they can not do intelligent processing lie upfront judgment, forecasting, analyzing data and mae positive efforts to maintain quality of service. These problems have motivated a trend towards distributed management intelligence that represents a rational approach to overcome the limitations of centralized NM. As a result, several distributed management framewors have been proposed both by researchers and standardization bodies [6] [7]. However, these models are typically identified by static management components that cannot adapt to the evolving nature of today s networs, with rapidly changing traffic patterns and topology structures. Of-late, the Mobile Agent (MA) paradigm has emerged within the distributed computing field. The term MA refers to autonomous programs with the ability to move from host to host to resume or restart their execution and act on behalf of users towards the completion of a given Vol 1, No 2 (October 2010) IJoAT 285

3 tas [8]. One of the most popular topics in MA research community has been distributed NM, wherein MAs have been proposed as a means to balance the burden associated with the processing of management data and decrease the traffic associated with their transfers (data can be filtered at the source). Networ management based-on Mobile agent refers to equipping agents with networ management intelligence and allowing them to issue requests to managed devices/objects after migrating close to them as depicted in figure 2. The independence and mobility of mobile agents reduce client server bandwidth problems by moving a query from client to the server. It not only saves repetitive request/response handshae but also addresses the much needed problems created by intermittent or unreliable networ connections. Agents can easily wor off-line and communicate their results when the application is bac on-line. Moreover, agents support parallel execution (load balancing) of large computation which can be easily divided among various computational resources Conventional Networ Management System The activities involving, operation, administration, maintenance and provisioning of networ resource and services, is called networ management [9]. Req - 1 MANAGER Res - 1 Req - n AGENT AGENT MANAGEMEN T APPLICATION S Client Res - n Server MANAGER MANAGEMEN T APPLICATION S Prepare & Dispatch MAs MAs return with mgmt data MA-Req-1 MA-Res-1 MA-Req-n MA-Res-n AGENT Figure 2 : Client/Server v/s Mobile Agents The relevant data from various agents, sitting on the networ devices, is gathered, analyzed and then employed to perform the appropriate management tass. The decisions thereof are taen either manually by the operators monitoring the networs or in an automated fashion by various management applications. Vol 1, No 2 (October 2010) IJoAT 286

4 A typical organization model of a networ management system is based on SNMP twotier architecture. It consists of two major components: networ agent process and the networ manager process. The networ agent process resides on the managed networ devices such as routers, switches, servers etc. The networ manager is housed on the NMS station from where it manages the various devices, by accessing the management information, through the agents residing on them as shown in fig. 3. The management information consists of collection of managed objects, stored in Management Information Base (MIB). Various management applications such as configuration management, fault management etc. resides on the NMS stations whereas manager/agent paradigm procures the needed data for respective management application. The agents have very simple interfaces by means of which they provide information to the requesting applications on granular basis gathered from the target devices. As they lac the needed intelligence and global view, agents don t perform management actions on their local data. Management data has to be transported to managers for taing any management decision. The management protocols provide the primitives for exchanging the management information among the agents and managers. Inherently it leads to centralized model of networ management. SNMP MANAGER SNMP MANAGER SNMP MANAGER SNMP AGENT SNMP AGENT NETWORK DEVICE NETWORK DEVICE MIB MIB Figure 3. Two-tier Manager/Agent models A centralized architecture suffers from the lac of scalability and flexibility. Furthermore, the staleness of gathered data (due to networ latency involved) and probable error in the selection of management tas being carried over (owing to the staleness of data) reduces the reliability of the management applications. These problems have motivated a trend towards distributed management intelligence that represents a rational approach to overcome the limitations of centralized NM. Gathering and analysis of the management data from agents of managed devices is partitioned and spread over the various computing platforms in the networ (sometime the managed devices act as computing platform) thereby breaing the centralized paradigm. Mobile agents allow more and more of the networ management intelligence to move closer to the devices unlie the centralized model. Some micromanagement operations could be performed locally avoiding the Vol 1, No 2 (October 2010) IJoAT 287

5 need to transfer large amount of data generated at management nodes to the central management station thereby reducing the worload for the management station and the overhead in the networ. 2. Related Literature The disadvantages of a totally centralized architecture as described before lead researchers and industry experts to loo for a form of decentralization in the networ management system design and architecture. In this context, the use of approaches based on mobile agents for networ management allows to overcome many limitations of current centralized management systems. Early wor in the field of mobile code, carried over by Goldszmidit et al [6], introduces the concept of management by delegation. Herein, the management station can extend the capability of the agents at runtime thereby invoing new services and dynamically extending the ones present in the agent on the device. Further wor in the field of mobility of code established three major paradigms, a. Code on demand, b. Remote evaluation, c. Mobile agents, introduced by Baldi et al[10]. Mobile agent based strategies have distinct advantages over the others as it allowed for easy programmability of remote nodes by migrating and transferring functionality wherever it is required. Bellavista et al.[11] proposed a secure and open mobile agent environment, MAMAS (Mobile Agents for the Management of Applications and Systems) for the management of networs, services and systems. Sahai & Morin [12] introduce the concept of mobile networ managers (MNM), which is a location independent networ manager and assists the administrator to remotely control his/her managed networ, through launching MAs to carry out distributed management tass. In [13], Oliveira and Lopes propose how the integration of MA-based sub-system could be carried out in the IETF s DISMAN framewor. Manoj Kumar Kona et al. [14] described an SNMP based efficient mobile agent networ management structure, in order to cooperate with conventional management system; For transferring less networ monitoring data and managing devices more effectively, Damianos Gavalas et al.[15] adopted the methods of calculating Health Functions of MIB variables and polling SNMP tables to deploy management agents; Chi-Yu Huang et al. proposed a clustering mobile agent based networ management model aiming at large enterprise entrant networ in [16]. Liotta et al. [17] have suggested an MA-based hierarchical and dynamic management architecture which deploys static middle managers who in turn can launch MAs. Pualiafito et al. [18] introduce the Mobile Agent Platform (MAP), used for monitoring the systems state by calculating aggregation functions combining several MIB values (health functions). Damianos Gavalas et al.[19] proposed a hierarchical and scalable management model where middle managers are themselves mobile and based on certain policies they dynamically segment the networ and deploy other mobile middle managers for data collection. Vol 1, No 2 (October 2010) IJoAT 288

6 2.1. Mobile Agents Based Networ Management Model Damianos Gavalas et al. [19] have discussed number of mobile agent-based networ management models in their research wor. Few important ones have been discussed and compared against certain design parameters in this section. 1. Flat Bed Model: For a particular management tas a single mobile agent is launched from a management station which then traverses the networ topology in a sequential manner, visiting each managed device and carrying out the assigned tas. Though the model relives the networ from the flood of request/response messages, it introduces the issue of roundtrip delays as the networ size grows. This leads to scalability issue if data has to be collected very frequently from managed devices. The size of MAs grows considerably in large networs. The model is shown in Figure 4. Management Station (Manager) : Mobile Agents : LAN Segment : Managed Device Figure 4: Flat Bed Model 2. Segmentation Model: Here the scalability issue is addressed by partitioning the networ into many administrative or geographical domains and assigning a single MA entity to each one of them. This brings high degree of parallelism in the data collection architecture and brings the response time down by many folds. The model is shown in Figure 5. Management Station (Manager) : Mobile Agents : LAN Segment : Managed Device Figure 5: Segmentation Model Vol 1, No 2 (October 2010) IJoAT 289

7 3. Hierarchical Static Middle Managers Model: The scalability problem is more adequately addressed by deploying hierarchical models wherein NM tass are delegated to MAs. They migrate to remote subnetwors/domains where they act as local managers and taes over the responsibility of local devices from the central manager. These models suffer from automatic adaptation of management system to changing networ configurations, i.e. mid level manager do not change the location where they execute. The model is shown in Figure 6. Management Station (Manager) : Mobile Agents : LAN Segment : Managed Device :Static MM Figure 6: Hierarchical Static Middle Managers Model 4. Hierarchical Mobile Middle Managers Model: In search of more flexible solutions, a concept of Mobile Middle Manager (MDM), referring to a management component that operates at an intermediary level between the manager and the management end points, is introduced. The mobility feature of the MDMs allows the management system to adapt dynamically to a changing networ conditions. MDMs can be deployed to or removed from a given networ segment in response to change in networ traffic or move to a least loaded host to optimize local resource usage. The model is shown in Figure 7. S Management Station (Manager) : Mobile Agents : LAN Segment :Mobile MM : Managed Device Figure 7: Hierarchical Mobile Middle Managers Model Vol 1, No 2 (October 2010) IJoAT 290

8 Design Parameters v/s Models International Journal of Advancements in Technology ISSN Comparison Analysis of C/S and Various MA-Based Management Models All the MA-based management models enumerated in the previous section and conventional client/server model (SNMP client/server model) are compared for five typical design parameters, i.e. scalability of management functions as the networ grows, load balancing amongst participating entities, fault-tolerance of the management model, dynamic adaptation to the changing needs of networ topology & traffic patterns and easy integration of new management services in the management model. Table 1 summarizes the comparative study. Table 1: Comparative Analysis of various Networ Management Model C/S Model FBM SM H-SMM H-MMM Scalability Poor. The networ gets flooded with management request/response messages. Becomes poor with increase in the size of networ and if data has to collected very frequently. Becomes poor with increase in the number of segments. Better than the C/S, FBM & SM models as Static Manager taes over the management tass from the central manager. Better than the C/S, FBM & SM models as Mobile Manager taes over the management tass from the central manager. Load Balancing Low as every management decision is taen on central manager. Management tass could be downloaded to mobile agents for which they can utilize the local resources available on the managed devices. Management tass could be downloaded to mobile agents for which they can utilize the local resources available on the managed devices. Higher than FBM & SM as static manager can relieve the central manager from mobile agents generation and they intern can utilize local resources more efficiently. Highest as Mobile middle manager itself can re-locate itself as the load on the host (on which it is running) increases. Fault- Tolerance Once the lin between a subnetwor and central manager breas, that subnetwor can t be managed Mobile Agents, once dispatched, can carryout the management tas manage the assigned even if the lin between them and central managers breas. Mobile Agents, once dispatched, can carryout the management tas manage the assigned even if the lin between them and central managers breas. Static Managers can manage their own subnetwors even if the lin between them and central managers breas. Mobile Managers can manage their own subnetwors even if the lin between them and central managers breas. Dynamic Adaptation Integration of new services No No No No Yes No Yes Yes Yes Yes Vol 1, No 2 (October 2010) IJoAT 291

9 3. Role Of Mobile Agents In Distribution Of Managemet Tass 3.1 Mobile Agent Technology Mobile agents were introduced in the early 90 s within the artificial intelligence research community, as semi-intelligent computer programs that assist a user with large amounts of complex information within a networ environment. These are typically dispatched from one node in a networ and transported to a remote node for execution. This concept of remote programming using mobile agents is considered as an alternative to the traditional client-server programming based on the remote procedure call or the static distributed object paradigm (e.g. CORBA). The primary goal of using mobile agents in management of telecommunication networ is reducing networ traffic by using load balancing and building scalable and reliable distributed networ management system. Some of the advantages of using agent technology in telecommunication networs are as identified and widely supported by a large number of researchers [20] are described below: 1. Reduce management data overflow. For very large amount of management data ept at remote hosts, it would be wise to dispatch a management entity to the managed hosts for taing the decisions rather than transporting all of the data to a central location and then processing it. To avoid overloading the networ the data should be processed in its locality rather than transferred over the networ. Mobile agents help decentralize management system by allowing independent remote programming. 2. Address networ latency. Modern networs are quite big in size and spread over large geographical areas with different sections connected through bac lin WAN channels. Various mission critical applications needing real time monitoring runs on these networs. Size of these networs doesn t mitigate the need to respond immediately to important events as a result of significant latencies involved. These can be unacceptable for some real-time systems or systems involved with on-line processing of information. Mobile agents offer a solution because they can be dispatched from a central controller to act locally and execute the controller s directions directly. 3. Adaptation of management protocols. In a large, distributed and heterogeneous networ it is a high lielihood that different devices/nodes adopt different management protocols (SNMP, TL1, CORBA etc.) as they come from different vendors and deploy different technologies. Mobile agents offer an excellent solution to be able to encapsulate these details from management stations and tal to visiting devices in their own native protocol language. 4. Off-line and independent execution. At times there are problems created by intermittent or unreliable networ connections between the networ management stations and managed devices. Agents can easily wor off-line and communicate their results when the application is bac on-line. 5. Dynamic adaptation. Mobile agents can sense their environment and react autonomously to changes. When a number of mobile agents are assigned to wor with a common goal they can distribute themselves among hosts in the networ to maintain the optimal Vol 1, No 2 (October 2010) IJoAT 292

10 configuration for solving the problem. In the case of a mobile agent moving across a number of host nodes, it can adapt its future behaviour according to information that it has already collected and stored in its state. 6. Dealing with heterogeneous environments. Networ computing is fundamentally heterogeneous, often from both hardware and software perspectives. Mobile agents are only dependent to their execution environment. Also, mobile agents are not coupled with other objects; instead they collaborate with other local agents and use the functionality they discover in each host. 7. Networ fault tolerance. In order to perform the above properties, agents must communicate to find their peers, to cooperate and negotiate in open environments. It is significant that agent systems build on an interface with a diversity of existing and upcoming standards at the underlying networ systems level. 3.2 Distribution of Networ Management Tass Typical Networ management systems have to deal with various functional areas in order to ensure availability of resources and services of the networ to its end users. SNMP and CMIP require gathering and analyzing large amount of data from the networ for various management tass. SNMP manages and monitors only networ elements and agents provide a limited and fixed set of functions [21]. Existing management models traditionally adopt a centralized, client/server (CS) approach wherein the management application (with the help of a manager), acting as clients, periodically accesses the data collected by a set of software modules (agents) placed on networ devices by using an appropriate protocol. These systems regularly suffer from poor scalability due to an increase in the amount of communication and generate too much traffic in the networ and the number of failures in nodes and channel. Therefore, there is a need to employ mobile agents as an autonomous entity in networ management and transfer the administration tass to them. Also under this situation the networ management tass and computational load are distributed instead of being centralized towards and on the manager host. One of the important goals of the networ management is to have balanced loading and reliable loading on the networ such that connections in the networ can be established quicly without noise, or several trails. Networ management also aims to organize the networs in order to wor professionally, successfully adjust to changes, and react to problems such as traffic patterns. The OSI management model recognizes the following important networ management functional areas. These are: Fault Management, Accounting Management, Configuration Management, Performance Management and Security Management. Besides these networs topology discovery is one of the fundamental functions of the management systems. Typically it involves finding the devices of the networ and interconnection among those. A more detailed implementation would focus on construction of more detailed views that may include, for example, services available on a device or devices that satisfy certain constraints. If the Vol 1, No 2 (October 2010) IJoAT 293

11 constraints are functions of device status, then we approach problem discovery. As the complexity of discovery grows, it is harder to implement using classical client/server approaches. The same principles as we saw in networ modeling can be used to diagnose networ faults. Detection of faults is a process of building a specialized model of the networ. For example, a simple agent performing selective discovery of nodes with utilization that exceeds a certain threshold builds a model of over-utilized nodes. If the constraints on discovery describe violations of what is considered normal behavior of networ elements, then the agents testing the constraints perform a fault detection function. Another area of importance is performance measurement, which involves gathering statistical information about networ traffic, methods to reduce, and present data. Measuring performance of networs using centralized SNMP based management is very difficult due to reasons lie networ delays and information traffic jam at the central management station. It is now widely recognized that the use of decentralization in this ind of applications potentially solves most of the problems that exist in centralized client/server solutions. Hence applications can be more scalable, more robust, can be easily upgraded or customized and they reduce the traffic in the networ. In a distributed networ, the networ operator monitors the trend of networ flow to assess networ performance and identify unusual conditions. The analysis of data can be achieved from the management information base. The management information base preserves various data objects for networ management. The information in management information base is ordered in clusters and maintained in a tree-lie structure. Thus management information base manage the complex networ tass in the distributed networ management environment. 4. Conclusion Mobile agents offer an easy re-configurable, flexible and scalable solution to the management of today s complex telecommunication networs thereby reduces the number of necessary human interactions. Many of the complex management tass can be delegated to agents whom agents can easily carry out without much intervention from the higher management layers. As discussed, the independence and mobility of mobile agents reduce bandwidth overloading problems by moving a processing of the management data and decision maing from centralized management stations to the managed devices thereby saving many repetitive request/response roundtrips and also address the problems created by intermittent or unreliable networ connections between the networ management stations and managed devices. Agents can easily wor off-line and communicate their results when the application is bac on-line. Moreover agents support parallel execution (load balancing) of large computation which can be easily divided among various computational resources. Thus using agents networ monitoring and other management tass can be easily decentralized. Vol 1, No 2 (October 2010) IJoAT 294

12 References [1] Agent based computing: A boolet for executing, [2] Stallings, W., SNMP and SNMPv2: The Infrastructure for Networ Management, IEEE Communications Magazine, vol.36, no.3, pp , March [3] Yemini, Y., The OSI Networ Management Model, IEEE Communications Magazine, vol.31, no.5, pp May [4] Liotta, A., Pavlou, G. Knight, G., Exploiting agent mobility for large-scale networ monitoring, IEEE Networs, 16:7-15, [5] Rubinstein, M.G., Duarte, O., Pujolie, G., Scalability of a networ management application based on mobile agents, Journal of Communication and Networs, 5: , [6] Goldszmidt G., Yemini Y., Yemini S., Networ management by delegation, Proceedings of the 2 nd Int. Symposium on Integrated Networ Management (ISINM 91), April [7] SNMP research, The Mid-Level Manager, [8] M. Baldi, S. Gai, G.P. Picco, Exploiting Code Mobility in Decentralized and Flexible Networ Management, in Proceedings of First International Worshop on Mobile Agents (MA 97), pp , Berlin, Germany, [9] Subramanian, M., Networ Management Principles and Practice, Addison Wesley, [10] Baldi M., Picco G.P., Evaluating the Tradeoffs of Mobile Code Design Paradigms in Networ Management Applications, Proceedings of the 20th Int. Conf. on Software Engineering (ICSE 98), pp , April1998. [11] Bellavista, P., Corradi A., STefanelli C., An open secure mobile agent framewor for systems management. Journal of Networ and System management, 7: , [12] Sahai, C. Morin, Towards Distributed and Dynamic Networ Management, in the Proceedings of IFIP/ IEEE Networ Operations and Management Symposium (NOMS), New Orleans, U.S.A., Feb, [13] Oliveira J.L., Lopes R.P., Distributed Management Based on Mobile Agents, Proceedings of the 1 st Int. Worshop on Mobile Agents For Telecommunication Applications, pp , Oct [14] M. K. Kona, C. Z. Xu, A Framewor for Networ Management Using Mobile Agents, Parallel and Distributed Processing Symposium, Proceedings International, pp , IPDPS [15] M. Ghanbari, D. Gavalas, D. Greenwood, M. O Mahony, Advanced networ monitoring applications based on mobile/intelligent agent technology, Computer Communications Journal, 23(8), pp , April [16] Chi-Yu Huang, Pattinson C., Using Mobile Agent Techniques for Distributed Manufacturing Networ Management, PGNet 2001, Liverpool, UK. [17] Liotta A., Knight G., Pavlou G., Modelling Networ and System Monitoring Over the Internet with Mobile Agents, Proceedings of the IEEE/IFIP Networ Operations and Management Symposium (NOMS 98), pp , February [18] Pualiafito A., Tomarchio O., Vita L., MAP: Design and Implementation of a Mobile Agents and Platform,Journal of Systems Architecture, 46(2), pp , January 2000 [19] Gavalas, D., D. Greenwood, M. Ghanbari and M. O Mahony, Hierarchical networ management: A scalable and dynamic mobile agent-based approach. Computer Networs, 38: , [20] C. Bohoris, G. Pavlou, H. Cruicshan, Using Mobile Agents for Networ Performance Management, in proceedings of the IFIP/ IEEE, Networ Operations and Management Symposium (NOMS 00), Hawaii, U.S.A., April [21] Tripathi, A., Ahmed, T., Patha, S., Paradigms for Mobile Agent Based Active Monitoring Of Networ Systems, IEEE, Vol 1, No 2 (October 2010) IJoAT 295